diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp index 233fb15c0a..4cc62509d7 100644 --- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp +++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp @@ -72,8 +72,8 @@ bool profile_gemm_splitk_impl(int do_verification, { case 0: break; case 1: - a_m_k.GenerateTensorValue(GeneratorTensor_2{0, 1}); - b_k_n.GenerateTensorValue(GeneratorTensor_2{-1, 1}); + a_m_k.GenerateTensorValue(GeneratorTensor_2{-1, 2}); + b_k_n.GenerateTensorValue(GeneratorTensor_2{-1, 2}); break; default: a_m_k.GenerateTensorValue(GeneratorTensor_3{0.0, 1.0}); @@ -94,7 +94,7 @@ bool profile_gemm_splitk_impl(int do_verification, a_device_buf.ToDevice(a_m_k.mData.data()); b_device_buf.ToDevice(b_k_n.mData.data()); - c_device_buf.ToDevice(c_m_n_device_result.mData.data()); + c_device_buf.SetZero(); using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK